A multi-agent reinforcement learning using Actor-Critic methods

This paper investigates a new algorithm in Multi-agent Reinforcement Learning. We propose a multi-agent learning algorithm that is extend single agent Actor-Critic methods to the multi-agent setting. To realize the algorithm, we introduced the value of agent's temporal best-response strategy instead of the value of an equilibria. So, our algorithm uses the linear programming to compute Q values. When there are multi Nash equilibrium in the games, the mixed equilibrium was be reached. Our learning algorithm works within very general framework of n-player, general-sum stochastic games, and learns both the game structure and its associated optimal policy.